┌─────────────────────────────────────────────────────────────────┐
│ B.Tech CSE – AI & ML │ CGPA 9.12 │ Applied AI │ Open to Internships 2026 │
└─────────────────────────────────────────────────────────────────┘
|
I'm an AI/ML engineer who builds things that actually work — from production ML pipelines serving real bank data to agentic AI systems that won national hackathons. With 2 years of hands-on applied AI experience, I've shipped systems across FinTech, AgriTech, GovTech, and Dairy operations — not just notebooks, but deployed, running software solving real problems. I'm looking for an internship where I can contribute from day one on real ML or data engineering problems. |
class Vedika:
role = "AI/ML Engineer"
exp = "2 years Applied AI"
cgpa = 9.12
strengths = [
"End-to-end ML pipelines",
"LLM Agents & RAG systems",
"FastAPI + Docker deployment",
"Data pipelines & analytics",
]
status = "📬 Open to internships" |
|
🥈 2nd Place India AI Impact GenAI Hackathon (IISc × IBM) |
🏅 Top 10 AASHA AI Hackathon 2025 |
📄 Published ML Research Agriculture Domain |
🚀 Deployed 3 live production apps & APIs |
Not side projects. Real systems, real users, real impact.
2nd Place — India AI Impact GenAI Hackathon (IISc × IBM)
An agentic AI system that acts as a Chartered Accountant for Indian MSMEs and startups — automating bookkeeping, GST compliance, financial reports, and offering a multilingual chatbot in Hindi + English.
IBM WatsonX Orchestrate · IBM Cloud · MongoDB · Custom Tool Packages
What makes it different: Built using IBM's Agent Development Kit with 5 modular Python tools (bookkeeping, reminders, report generation, financial calculator, multilingual chat) — all orchestrated through WatsonX. Not a chatbot wrapper — a real agentic workflow.
End-to-end production ML pipeline for retail banking churn prediction on 50,000 customer records — from EDA notebooks to a live deployed FastAPI inference endpoint.
LightGBM · SHAP Explainability · FastAPI · Docker · Scikit-learn Pipeline · Render
| Metric | Value |
|---|---|
| Dataset | 50,000 records · 20 features · 17.4% churn rate |
| Model | LightGBM + sklearn Pipeline (no training/serving skew) |
| Test ROC-AUC | 0.7866 |
| Explainability | SHAP TreeExplainer — top-5 feature contributions per prediction |
| Deployment | Docker → Render · Live Swagger docs |
What makes it different: Not just a model — a full ML system. Includes threshold tuning for business actionability (0.61), SHAP-based local explanations in the API response, and reproducible training scripts separate from notebooks.
ML crop recommendation system for farmers — predicts the top 3 suitable crops from soil and climate inputs, pulls live market prices from a government API, and answers farming questions via Gemini AI.
Random Forest · Gemini API · Streamlit · data.gov.in API · Cosine Similarity
| Feature | Detail |
|---|---|
| ML Accuracy | 87%+ on held-out test data |
| Input Features | N, P, K, pH, Rainfall, Temperature |
| Market Data | Live prices from data.gov.in · offline fallback |
| AI Layer | Gemini — crop alternatives via cosine similarity + justifications |
What makes it different: Combines an ML model + live government API + Gemini in one pipeline. The profit calculator and alternative crop engine make it genuinely useful for real farmers, not just a demo.
Top 10 — AASHA Hackathon 2025
AI-powered career assistant built specifically for women — empathetic conversational AI for job discovery, internship guidance, skill-gap analysis, and women-centric government scheme information.
Gemini API · Streamlit · Python · Lottie Animations · Ethical AI Bias Detection
What makes it different: Has an ethical AI layer that detects and restricts biased or discriminatory inputs. Session-based memory keeps conversations coherent. Built for real social impact at a national hackathon.
Real-world vehicle scheduling system built for and deployed at Krushna Dudh Sangh, Islampur — a dairy cooperative that was manually managing their fleet of delivery vehicles on paper and Excel.
Python · Pandas · Streamlit · PDF Generation · Conflict Detection Logic
| Before | After |
|---|---|
| 3–4 hours of manual daily planning | Minutes — fully automated |
| Frequent route conflicts & duplicates | Zero conflicts — built-in detection |
| Paper schedules, hard to update | Downloadable PDF timetables — colour coded |
What makes it different: This is a real client project — built in collaboration with actual stakeholders, validated against their operational requirements, and running in production today.
AI-assisted cybersecurity policy generation tool — users input system context, the app generates a structured security policy using a rule engine + local Ollama LLM chatbot for Q&A. Built during NPTEL Cybersecurity course.
FastAPI · Ollama (Llama 3.2) · Jinja2 · HTML/CSS · PDF Export · Uvicorn
What makes it different: Demonstrates backend API design, local LLM integration (privacy-first, no external API calls), Jinja2 templating, and PDF export — a full-stack mini product, not just a script.
Languages
ML & Data
LLMs & Agents
Backend & Deploy
Databases
If you're working on interesting ML, data, or applied AI problems — I'd love to connect.
